[LLVMbugs] [Bug 23376] New: AVX512: We don't always need EVEX for VEX-encodable instructions.

[bugzilla-daemon at llvm.org]

https://llvm.org/bugs/show_bug.cgi?id=23376

            Bug ID: 23376
           Summary: AVX512: We don't always need EVEX for VEX-encodable
                    instructions.
           Product: libraries
           Version: trunk
          Hardware: PC
                OS: All
            Status: NEW
          Severity: normal
          Priority: P
         Component: Backend: X86
          Assignee: unassignedbugs at nondot.org
          Reporter: ahmed.bougacha at gmail.com
                CC: llvmbugs at cs.uiuc.edu
    Classification: Unclassified

This:

define float @t(float %a, float %b) {
  %res = fadd float %a, %b
  ret float %res
}

Generates the EVEX-prefixed:

    62f17e0858c1        vaddss    %xmm1, %xmm0, %xmm0

instead of using VEX:

    c5fa58c1            vaddss    %xmm1, %xmm0, %xmm0


AFAIK, there is no transition penalty between AVX512 and AVX, so I believe the
only reason to use EVEX in this case is to access more registers (xmm0-31).

That's not always useful, so we should recognize that, and relax EVEX->VEX when
possible.

-- 
You are receiving this mail because:
You are on the CC list for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-bugs/attachments/20150430/b0adda19/attachment.html>